<![CDATA[Fog phenomena pose a significant challenge in aviation operations, particularly in regions with complex topography such as Ruteng Airport. Thick fog can reduce visibility and increase flight safety risks. This study aims to develop a deep learning-based fog prediction model by comparing the performance of Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) enhanced with Attention Mechanism (AM). The dataset consists of 61,187 entries, covering hourly recorded weather parameters over the past ten years (2013–2023). The experimental results show that the addition of Attention significantly improves model performance. The RNN+Attention model emerges as the best-performing model with an accuracy of 0.9981, precision of 0.7755, recall of 0.76, and F1-score of 0.7677, along with the lowest number of False Positives. Meanwhile, the LSTM+Attention model excels in reducing False Negatives, making it suitable for systems prioritizing comprehensive fog detection. Models without Attention demonstrate perfect recall (1.00), but their low precision indicates overfitting. Overall, the integration of the Attention Mechanism enhances the balance between recall and precision and improves model reliability in handling data imbalance. The contribution of this research is that it can serve as a reference for future studies in developing artificial intelligence-based weather prediction models, particularly in addressing fog phenomena. Keywords – Attention Mechanism; Long Short-Term Memory; Fog Prediction; Recurrent Neural Network]]>
Copyrights © 2025
